Re: [Numpy-discussion] A proposal for dtype/dtypedescr in numpy objects